1.root账号先在namenode节点上配置一个定时任务,将fsimage定时传到其他客户机上进行操作
whereis hadoop命令确定安装目录,然后去配置文件找到namenode节点(data-93 emr-header-1)
0 1 * * * sh /root/fsimage.sh 每晚一点将fsimage文件发送到集群其他机器上,fsimage.sh如下
#!/bin/bash
TARGET_HOST=192.168.11.130
SCP_PORT=
IMAGE_DIR=/mnt/disk1/hdfs/name/current
TARGET_DIR=/data/hdfs
DAY=`date +%Y%m%d`
echo "day=$DAY"
cd $IMAGE_DIR
fsname=`ls fsimage* | head -`
echo $fsname
scp -P $SCP_PORT $fsname ${TARGET_HOST}:${TARGET_DIR}/fsimage.${DAY}
echo "done"
脚本在/mnt/disk1/hdfs/name/current下执行【scp -P 58422 fsimage_0000000007741688997 192.168.11.130:/data/hdfs/fsimage.20190920】,将namenode上的fsimage镜像文件传递到data130(192.168.11.130)上的文件夹里
2.切换账号gobblin,在data-130的机子上配置crontab 任务,每天2点执行分析脚本
small_file_analysis.sh如下
#!/bin/bash
source /etc/profile
basepath=$(cd `dirname $`; pwd)
cd $basepath
IMAGE_DIR="/data/hdfs"
IMAGE_PREFIX="fsimage"
cur_date="`date +%Y%m%d`"
cur_month="`date +%Y%m`"
cur_day="`date +%d`"
echo "cur month = $cur_month"
echo "cur day = $cur_day"
echo "cur date = $cur_date"
IMAGE_NAME=$IMAGE_PREFIX.$cur_date
echo "fsimage name is $IMAGE_NAME"
export HADOOP_HEAPSIZE=
hdfs oiv -i $IMAGE_NAME -o $IMAGE_NAME.txt -p Delimited
hive -e "load data local inpath '$IMAGE_DIR/$IMAGE_NAME.txt' overwrite into table dataplatform.fsimage partition (month='$cur_month',day='$cur_day');"
hive -hivevar CUR_MONTH=$cur_month -hivevar CUR_DAY=$cur_day -f small_file_analysis.hql
rm -f fsimage*
echo "done"
脚本逻辑很简单:使用image分析工具iov将image转为txt格式的文件,然后将文件导入hive 表(dataplatform.fsimage),再通过hive命令执行sql,将sql查询结果插入分析结果表(dataplatform.small_file_report_day),最后删除fsimage开头的2个文件即可
注意:export HADOOP_HEAPSIZE=10240 要加上,不然会报堆内存溢出
设置堆内存大小之后执行:
small_file_analysis.hql 如下:
set mapreduce.job.queuename=root.production.gobblin;
set mapreduce.job.name=small_file_analysis;
set hive.exec.parallel=true;
set hive.exec.parallel.thread.number=4;
set mapreduce.map.memory.mb=1024;
set mapreduce.reduce.memory.mb=1024;
INSERT OVERWRITE TABLE dataplatform.small_file_report_day PARTITION (month='${CUR_MONTH}', day='${CUR_DAY}')
SELECT b.path as path, b.total_num as total_num FROM (
SELECT path, total_num, root_path
FROM
(
SELECT
SUBSTRING_INDEX(path, '/', 4) AS path,
COUNT(1) AS total_num,
SUBSTRING_INDEX(path, '/', 2) AS root_path
FROM
dataplatform.fsimage
WHERE
file_size < 1048576
AND month='${CUR_MONTH}' AND day='${CUR_DAY}'
AND SUBSTRING_INDEX(path, '/', 2) in ('/warehouse', '/tmp')
GROUP BY SUBSTRING_INDEX(path, '/', 4),SUBSTRING_INDEX(path, '/', 2)
UNION
SELECT
SUBSTRING_INDEX(path, '/', 5) AS path,
COUNT(1) as total_num,
SUBSTRING_INDEX(path, '/', 3) AS root_path
FROM
dataplatform.fsimage
WHERE
file_size < 1048576
AND month='${CUR_MONTH}' AND day='${CUR_DAY}'
AND SUBSTRING_INDEX(path, '/', 3) = '/gobblin/source'
GROUP BY SUBSTRING_INDEX(path, '/', 5),SUBSTRING_INDEX(path, '/', 3)
) a
dataplatform.fsimage建表语句
CREATE TABLE `fsimage`(
`path` string,
`block_num` int,
`create_time` string,
`update_time` string,
`block_size` bigint,
`unknown1` int,
`file_size` bigint,
`unknown2` int,
`unknown3` int,
`permission` string,
`user` string,
`group` string)
PARTITIONED BY (
`month` string,
`day` string)
ROW FORMAT SERDE
'org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe'
WITH SERDEPROPERTIES (
'field.delim'='\t',
'serialization.format'='\t')
STORED AS INPUTFORMAT
'org.apache.hadoop.mapred.TextInputFormat'
OUTPUTFORMAT
'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION
'hdfs://emr-cluster/warehouse/dataplatform.db/fsimage'
dataplatform.small_file_report_day建表语句:
CREATE TABLE `dataplatform.small_file_report_day`(
`path` string,
`total_num` bigint)
PARTITIONED BY (
`month` string,
`day` string)
ROW FORMAT SERDE
'org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe'
STORED AS INPUTFORMAT
'org.apache.hadoop.hive.ql.io.parquet.MapredParquetInputFormat'
OUTPUTFORMAT
'org.apache.hadoop.hive.ql.io.parquet.MapredParquetOutputFormat'
LOCATION
'hdfs://emr-cluster/warehouse/dataplatform.db/small_file_report_day'
TBLPROPERTIES
'parquet.compression'='SNAPPY'
手机扫一扫
移动阅读更方便
你可能感兴趣的文章